> ## Documentation Index
> Fetch the complete documentation index at: https://anthale.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Secure AI Systems Guides for Agent, RAG, and MCP Security

> Choose the Anthale guide for AI agent security, RAG pipeline security, MCP integration security, tool call validation, or LLM output validation.

Use this section when you need a practical guide for AI agent security, RAG security, MCP security, browser-agent hardening, memory safety, tool call validation, or LLM output validation. Each page focuses on one implementation task and links back to the relevant Anthale guardrails and threat references.

## Start with the task you need to solve

<Columns cols={2}>
  <Card title="AI Agent Security" icon="bot" href="/docs/learn/secure-ai-systems/secure-agent-workflows">
    Map the boundaries across user input, retrieval, tools, memory, and model output.
  </Card>

  <Card title="Tool Call Validation" icon="wrench" href="/docs/learn/secure-ai-systems/gate-tool-actions">
    Keep model-proposed actions behind least privilege, validation, and approval gates.
  </Card>

  <Card title="RAG Security" icon="database" href="/docs/learn/secure-ai-systems/secure-retrieval-pipelines">
    Treat retrieved text as untrusted context until your application and policy say otherwise.
  </Card>

  <Card title="Browser Agent Security" icon="globe" href="/docs/learn/secure-ai-systems/secure-browser-agents">
    Keep web pages, DOM content, and browser actions from becoming an unsafe authority path.
  </Card>

  <Card title="LLM Output Validation" icon="file-check" href="/docs/learn/secure-ai-systems/validate-model-output">
    Validate the model's final response before it reaches a user, a tool, or storage.
  </Card>

  <Card title="AI Memory Security" icon="brain" href="/docs/learn/secure-ai-systems/secure-memory-and-state">
    Stop poisoned summaries or saved context from becoming a persistent attack path.
  </Card>

  <Card title="MCP Security" icon="plug" href="/docs/learn/secure-ai-systems/secure-mcp-integrations">
    Connect MCP servers with server scoping, tool review, and runtime controls in place.
  </Card>

  <Card title="AI Agent Red Teaming" icon="shield" href="/docs/learn/secure-ai-systems/red-team-agent-workflows">
    Test prompt injection, tool abuse, retrieval poisoning, and memory failures before they reach production.
  </Card>
</Columns>

## Use the guardrail references when you need the threat model

<Columns cols={2}>
  <Card title="Prompt Injection Prevention" icon="syringe" href="/docs/learn/guardrails/prompt-injection">
    Start with Anthale's core prompt injection control model.
  </Card>

  <Card title="Advanced Prompt Injection Paths" icon="route" href="/docs/learn/guardrails/prompt-injection/advanced-attack-paths">
    Expand beyond direct prompt attacks into memory, tool, MCP, and multimodal paths.
  </Card>

  <Card title="Prompt Injection Obfuscation" icon="binary" href="/docs/learn/guardrails/prompt-injection/obfuscation-evasion">
    See how encoding, wrappers, and fragmented payloads hide attacker intent.
  </Card>

  <Card title="Map Prompt Injection Attack Surfaces" icon="search-x" href="/docs/learn/guardrails/prompt-injection/attack-surfaces">
    Find the boundaries where untrusted text reaches model context.
  </Card>

  <Card title="Prompt Injection Impact Paths" icon="shield-alert" href="/docs/learn/guardrails/prompt-injection/impact-paths">
    Classify what an injection enables after it lands.
  </Card>

  <Card title="Data Exfiltration Patterns" icon="scan-search" href="/docs/learn/guardrails/data-leakage/exfiltration-patterns">
    Classify recurring leak attempts across model output, links, and tool-assisted retrieval.
  </Card>
</Columns>

## Apply this in Anthale

* [Create a first policy](/docs/quickstart/first-policy) when you want to turn one of these security tasks into a working Anthale enforcement path.
* [See the enforcement API](/docs/api-reference) when you need the exact runtime contract, authentication details, and error behavior.
* [Request access](https://anthale.com/#request-access) when you want to review a live workflow with Anthale in place.

## Suggested paths

* Building a tool-using agent: start with [Secure Agent Workflows](/docs/learn/secure-ai-systems/secure-agent-workflows), then [Gate Tool Actions](/docs/learn/secure-ai-systems/gate-tool-actions).
* Hardening a retrieval flow: start with [Secure Retrieval Pipelines](/docs/learn/secure-ai-systems/secure-retrieval-pipelines), then [Validate Model Output](/docs/learn/secure-ai-systems/validate-model-output).
* Hardening a browser-driven workflow: start with [Secure Browser Agents](/docs/learn/secure-ai-systems/secure-browser-agents), then [Gate Tool Actions](/docs/learn/secure-ai-systems/gate-tool-actions).
* Connecting AI clients and external servers: start with [Secure MCP Integrations](/docs/learn/secure-ai-systems/secure-mcp-integrations), then [Log Security Events](/docs/learn/secure-ai-systems/log-security-events).
* Building a security review loop: start with [Red-Team Agent Workflows](/docs/learn/secure-ai-systems/red-team-agent-workflows), then [Log Security Events](/docs/learn/secure-ai-systems/log-security-events).
* Reviewing implementation quality: start with [Review AI Security in Code](/docs/learn/secure-ai-systems/review-ai-security-in-code), then [Red-Team Agent Workflows](/docs/learn/secure-ai-systems/red-team-agent-workflows).

## Anthale-specific setup

Use [Documentation MCP](/docs/build-with-ai/mcps/documentation-mcp) when you want read-only Anthale documentation lookup inside an AI client. Use [Prompt Injection Hardening Skill](/docs/build-with-ai/agent-skills/prompt-injection-hardening) when you want a coding agent to review or harden code with prompt injection, tool, retrieval, and boundary risks in mind.

## Next steps

Continue with [Secure Agent Workflows](/docs/learn/secure-ai-systems/secure-agent-workflows) or [Secure AI Systems](/docs/secure-ai-systems).